A fault tolerant multi-sensor fusion navigation system for drone in urban environment

Abstract

Precise positioning becomes an attractive research area to enhance last-mile delivery with drones. However, the reliability of precise poisoning is significantly degraded in GNSS-denied environments such as urban canyons. In this case, the excellent performance of Visual Inertial Odometry (VIO) in local pose estimation makes visual navigation technology more feasible for researchers. However, the accuracy and robustness of VIO degrade in faulted conditions. This paper presents a fault-tolerant multisensor fusion navigation system for drones in urban environments. We first performed Failure Mode and Effect Analysis (FMEA) in the VIO system to identify potential failure mode, which is feature extraction errors. Then, an integrated, loosely coupled EKF-based VIO system is proposed for our GNSS/VINS/LIO reference system to mitigate visual and IMU faults. The performance of the proposed method was validated by a synthetic dataset created using MATLAB, and it has shown improved robustness over Visual odometry and state-of-art VINS systems.